Targeting resistant tumour areas produced only a small additional gain in predicted tumour control

Summary

• Targeting extra radiation at the most resistant parts of a lung tumour improved predicted tumour control compared with giving the tumour a standard uniform dose.

  
  

• But compared with the simpler option of increasing radiation across the whole tumour, the personalised approach improved predicted tumour control by only about 3 percentage points on average.

  
  

• Only about 2% of the simulated cases showed a substantially larger benefit from targeting resistant tumour areas, suggesting the more complex approach may offer limited additional value for most patients.

Why might different parts of a tumour need different radiation doses?

Radiotherapy damages cancer cells with ionising radiation. But a tumour is not always uniform: some regions may be more resistant to radiation than others.

Dose painting by numbers is an approach that uses medical imaging to estimate those differences and assign a higher dose to regions expected to be more resistant. The challenge is that the underlying maps are estimates. Patient movement, imaging error and changes during a course of treatment can all make the apparent target less certain.

Why is this a prescriptive analytics problem?

A treatment-planning system must decide how much radiation to deliver to different parts of a tumour while limiting unnecessary exposure elsewhere. When the biological information used to make that decision is uncertain, the plan also needs to remain effective across several plausible versions of the patient's tumour response.

This study used robust optimisation, which means designing a plan to perform well across a range of uncertainty rather than assuming one estimated map is exactly correct.

How did the team compare the treatment strategies?

Researchers from Amsterdam Business School, the University of Texas, Massachusetts General Hospital and Harvard Medical School developed an automated planning framework.

Using imaging scans from 12 lung cancer patients, they generated 324 scenarios representing different tumour sensitivities, radiation responses and possible imaging errors. They compared dose painting with nonselective dose escalation - raising the dose across the whole tumour - and with a standard uniform-dose approach.

Automating the planning process made the comparison more consistent because the alternatives were generated under the same rules rather than being adjusted differently by individual planners.

What did the findings show?

Dose painting produced a larger estimated improvement in tumour control than a standard uniform dose. But compared with the simpler strategy of increasing the dose across the whole tumour, the average additional gain was about 3 percentage points.

A meaningfully large advantage for dose painting appeared in only about 2% of the simulated cases. The study therefore suggests that, under the uncertainties tested, much of the potential benefit could be achieved by the simpler strategy for many patients.

These results come from treatment-planning simulations rather than a clinical trial, so they do not show that one strategy produces better survival or side-effect outcomes in practice. They do indicate where more complex imaging and planning may add enough value to justify further clinical evaluation.

How can the findings inform other applications?

The comparison provides a way to identify when additional treatment complexity is likely to change the decision and when a simpler strategy performs similarly. This principle is relevant to other forms of personalised radiotherapy in which imaging-based biological information is uncertain.

Future research could test the approach with clinical outcomes, other tumour types and imaging methods, and examine whether specific patient characteristics can identify the smaller group most likely to benefit from dose painting.